TBC2014 poster "Outcome-guided mutual information networks for investigating gene-gene interaction effects on clinical outcomes", Hyun-hwan Jeong, So Yeon Kim, Kyubum Wee, Kyung-Ah Sohn

1. Methods
Outcome-guided mutual information network construction
1) Integrative network construction
2) Network-based survival analysis
Outcome-guided mutual information networks for investigating gene-gene
interaction effects on clinical outcomes
Hyun-hwan Jeong, So Yeon Kim, Kyubum Wee, Kyung-Ah Sohn
Department of Information and Computer Engineering, Ajou University, Suwon 443-749, S. Korea
e-mail : {libe,jebi1771,kbwee,kasohn}@ajou.ac.kr
Introduction
Network-based analysis frameworks have gained huge popularity recently as network
information can provide a more systematic and global view of the underlying biological
system. However, most network-based studies rely on feature-wise networks which can
reveal the relation between a pair of features, but do not consider the effect of pair-wise
feature interactions on the outcome.
To detect significant feature pairs associated with the outcome, we employ the Mutual
Information measure, which is a non-parametric, information-theoretic measure and has
been successfully used to detect linear or non-linear association between the features. Based
on the extension of Mutual Information measure, we propose a simple but powerful scheme
to construct an outcome-guided network with appropriate edge significance filtering.
We demonstrate the utility of the proposed network construction approach in two main
applications: the integrative network analysis of multiple genomic profiles, and the
network-based survival analysis. In both applications, datasets from ovarian serous
cystadenocarcinoma patients in The Cancer Genome Atlas (TCGA) are used. The results
highlight the usefulness of the outcome-guided mutual information networks in both
applications for investigating gene-gene interaction effects associated with clinical
outcomes.
References
[1] Cerami, E., et al., The cBio Cancer Genomics Portal: An Open Platform for Exploring Multidimensional Cancer Genomics Data. Cancer Discovery, 2012. 2(5): p. 401-404.
[2] TCGA, Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature, 2008. 455(7216): p. 1061-1068.
[3] Steuer, R., et al., The mutual information: detecting and evaluating dependencies between variables. Bioinformatics, 2002. 18(suppl 2): p. S231-S240.
[4] Butte AJ, Kohane IS, Mutual information relevance networks: functional genomic clustering using pairwise entropy measurements. Pacific Symposium on Biocomputing Pacific Symposium on Biocomputing, 2000:418-429.
[5] Li C, Li H, Network-constrained regularization and variable selection for analysis of genomic data. Bioinformatics, 2008. 24(9):1175-1182.
Results
Empirical distribution of mutual information values
Heatmap for the regression coefficients of 15 selected genes
(1) Significance of outcome-guided mutual information values
penalty term
𝒍 𝒑𝒆𝒏 𝜷, 𝒉 𝟎 = 𝒍 𝜷, 𝒉 𝟎 −
𝟏
𝟐
𝝀𝜷′
[ 𝟏 − 𝜶 𝑳 + 𝜶𝑰]𝜷
Network constrained regularized Cox regression
𝐿 = 𝐼 − 𝑆
𝐼 : identity matrix
𝑆 : normalized Laplacian matrix
α ∈ (0,1] : parameter which
controls the contribution of
network information
Prediction accuracy of the mutual information network-based Net-Cox model
(2) Integrative network analysis
(3) Network-based survival analysis
Significant GO terms
Intersection-network of whole genomic profiles
Category Description p-value FDR
BP hemopoiesis 1.82E-05 6.81E-03
BP immune system development 4.12E-05 6.81E-03
BP aging 3.03E-04 1.36E-02
BP T cell differentiation 4.69E-04 1.99E-02
BP positive regulation of apoptotic process 7.47E-04 2.02E-02
BP apoptotic process 5.92E-04 2.02E-02
BP placenta development 1.07E-03 2.44E-02
BP positive regulation of T cell activation 1.08E-03 2.44E-02
BP signal transduction by phosphorylation 1.49E-03 2.90E-02
BP cellular response to abiotic stimulus 1.68E-03 2.93E-02
Networks for single profile
G1 G2 …
Survival
month
0.5 -0.7 ... 15.0
1.0 0.4 ... 46.0
... ... ... ...
Integrative networks
a binary clinical outcome
discrete genomic profiles
𝐼(𝑋1, 𝑋2; 𝑌) = 𝐻(𝑋1, 𝑋2) + 𝐻(𝑌) − 𝐻(𝑋1, 𝑋2, 𝑌)
Mutual information(M.I.)
Statistically significant
gene pair
gene
Extraction
Gene pairs using
Mutual Information
Single profile networks
Integration
scheme
Outcome-guided
mutual information network